Professional Documents
Culture Documents
Hua Wang
School of Electrical and Computer Engineering
Georgia Institute of Technology, USA
hua.wang@ece.gatech.edu
Thermal Management,
DC IR Drop and Battery Life, etc…
Reliability
𝑷𝑨 𝑷𝑨𝑬 𝜼𝑫𝒆𝒗𝒊𝒄𝒆 𝜼𝑷𝑨 𝑴𝒐𝒅𝒆 𝜼𝑷𝑨 𝑮𝒂𝒊𝒏 𝜼𝑶𝒖𝒕𝒑𝒖𝒕 𝑳𝒐𝒔𝒔 𝜼𝑻𝒉𝒆𝒓𝒎𝒂𝒍 𝑨𝒈𝒊𝒏𝒈
~ ZOut/(ZOut+ZL)
-20dB/dec
~ Johnson’s Limit
= (VDD-VKnee)2/2RL
Larger devices or more devices
Lossy impedance transformation
π/2-BPSK 30%
QPSK 17.5%
16 QAM 12.5%
64 QAM 8%
256 QAM 3.5%
Need: High Linearity and
High Data Rate PAs
Carrier Output
Frequency Power
Energy Spectrum
Efficiency Efficiency
IL = 500mA RL = R S =
@ Imax = 1A 100ohm
IL = 20mA
VL
Vmax = 2V 50V
20mW 12.5W
Hua Wang Fundamentals of RF and Mm-Wave 23 of 98
Power Amplifier Designs
Conjugate Matching vs. Loadline Matching (3 of 3)
YS = 1/ZS IL
Example of a typical silicon power device
0 < IS < Imax
Imax=1A, Rs=100 ohm, Vmax=2V
IS VL = VS < Vmax YL = 1/ZL
Conjugate Matching
Desired RL = Rs = 100ohm, PL = 12.5W @ Imax = 1A
VL = Vs = 50V??? Device Breakdown 1W (Loadline)
IL
Now considering Vmax = 2V IL = Imax = 1A
RL = 2ohm = Vmax/Imax
IL = 20mA with IS = 40mA PL = 20mW in reality! << Rs = 100ohm
RL = Rs =
Loadline Matching IL = 500mA 100ohm
harmonics Id (DC)
VL
VDD 0 t
RF 0 t
Choke IDC = Id (DC)
IL Vd
Vd
Id VL
Vin VDD
RL
f0 = 1/√(LC) 0 t
Class A AB B C
θ 2ϖ ϖ<θ<2ϖ ϖ 0<θ<ϖ
1 𝜃 − 𝑠𝑖𝑛𝜃
𝜂= ·
4 sin 𝜃 − 𝜃 cos 𝜃
2 2 2
Id Ic VL
Vin IL (1) Let ω0t = θ. (2) Normalize
Vd = Vc I0 sin(ω0t+Φ) RL V and I to Class-B PA
Id Id, peak 1
𝑉 = 1 − 𝑐𝑜𝑠𝜃 − 𝑠𝑖𝑛𝜃+ 𝑠𝑖𝑛2𝜃 (1) Only for
Amplitude Id (f0) = Imax = Id, peak/2 2 waveform shaping
1
IDC = Id (DC) = Id, peak/pi 𝐼 = 1 𝑐𝑜𝑠𝜃 + 0 𝑐𝑜𝑠2𝜃 (2) No RF power
2
θ
0 π/2 3π/2 5π/2 Class-B PDC Class-B RF Pout [Wright09]
Hua Wang Fundamentals of RF and Mm-Wave 32 of 98
Power Amplifier Designs
Class-J and Class-J* PA
Generalization of Class-J PA and its continuum [Cripps09]
α = -1
𝟏 No RF power α = -0.5
Given 𝑰𝒅 = 𝟏 + 𝒄𝒐𝒔𝜽 + 𝟎 + 𝒄𝒐𝒔𝟐𝜽 α=0
𝟐 α = 0.5
α=1
𝟏
𝑽𝒅 = 𝟏 − 𝒄𝒐𝒔𝜽 − 𝜶𝒔𝒊𝒏𝜽 + 𝜶𝒔𝒊𝒏𝟐𝜽
𝟐
= 𝟏 − 𝒄𝒐𝒔𝜽 × 𝟏 − 𝜶𝒔𝒊𝒏𝜽 , −𝟏 < 𝜶 < 𝟏
Switching PAs
PA devices operating as switches
“On” as ideal short: can conduct high current + zero voltage
“Off” as ideal open: can withstand high voltage + zero current
Dedicated passive networks + device duty cycle Harmonics tuning and waveform
shaping
Vd Output power PL
OFF ON OFF ON
𝟏 𝟒 𝟏
Amplitude Vd (f0) Class-F 𝑷𝒐𝒖𝒕 = 𝑽 𝑰𝒎𝒂𝒙 > 𝑽𝑫𝑫 𝑰𝒎𝒂𝒙
VDD 𝟐 𝝅 𝑫𝑫 𝟐
= (4/pi)×VDD
Id Output power PL
Id, peak ON OFF ON OFF
𝟏 𝟒 𝝅
Amplitude Id (f0) Class-F-1 𝑷𝒐𝒖𝒕 = 𝑰 𝑽 = 𝑽𝑫𝑫 𝑰𝒎𝒂𝒙
𝟐 𝝅 𝒎𝒂𝒙 𝟐 𝑫𝑫
= (4/pi)×Id, peak/2 = (4/pi)×Imax
0 t Optimum load Ropt
Vd pi×VDD
pi×VDD 𝝅 𝟒
Class-F-1 𝑹𝒐𝒑𝒕 = 𝑽 / 𝑰 > 𝑽𝑫𝑫 /𝑰𝒎𝒂𝒙
Amplitude Vd (f0) 𝟐 𝑫𝑫 𝝅 𝒎𝒂𝒙
VDD = pi×VDD/2
0 t Drain efficiency η
Vd(t) and Id(t) waveforms have no overlap!
Zero power wasted in the device ηF-1=100%
VDD
RF f0=1/√(LC) with high Q Ic(t)
choke
I(t) I(t)=IDC+I0 cos(ω0t)
I(t) IDC
Id = ISW I0
Ic VL
IL IDC
Vd = Vc I0 cos(ω0t) RL t t
Cd -T1 0 T1 T0 -T1 -T1 0 T* T0 -T1
Id(t) = ISW(t)
ON ON
OFF OFF
t
-T1 0 T* T0 -T1
VDD
RF f0=1/√(LC) with high Q
choke
I(t) IDC Ic(t)
Id = ISW I(t) I(t)=IDC+I0 cos(ω0t)
Ic VL A1
IL
Vd = Vc
I0
Cd I0 cos(ω0t) RL
IDC A2
t t
-T1 0 T1 T0 -T1 -T1 0 T* T0 -T1
T1
Id(t) = ISW(t)
Vd(t) Vd(t) peak
ON ON
OFF OFF
Charging Discharging
t t
-T1 0 T* T0 -T1 -T1 0 T* T1 T0 -T1
I0
0 t t
0
Series LC resonance
0 t VA(t)
V0
VSWA VSWB IB LC parallel resonance
OFF ON OFF ON t
2IDC 0 VA or VB =0 before
SWA IA IB SWB
VB(t) SWA or SWB are on.
Vin Vin V0
ZVS
0 t
t
0
Hua Wang Fundamentals of RF and Mm-Wave 43 of 98
Power Amplifier Designs
Outline
Background and Motivations
VDD
RF 𝑉 = 𝑍 𝐼 = 𝑅 +𝑋 ·𝐼 ≤𝑉 ZL = RL + j√1-K2 × Ropt
Choke IL = Imax jXL
Vd ZL = R L Ropt = 12.5Ω
for Max Pout
Id V (f ) ≤ V
Vin d 0 DD
RL 𝑋 ≤ 𝑅 −𝑅 = 1−𝐾 ·𝑅 K × Max Pout
ZL = RL - j√1-K2 × Ropt
VDD
RF
𝐼 = 𝑌 𝑉 = 𝐺 +𝐵 ·𝑉 ≤𝐼 YL = GL - j√1-K2 × Gopt
Choke IL (f0) ≤ Imax
Ropt = 12.5Ω
Vd for Max Pout
Id
Vin Vd = VDD
jBL GL=1/RL 𝐵 ≤ 𝐺 −𝐺 = 1−𝐾 ·𝐺 K × Max Pout YL = GL = 1/RL
YL = GL + j√1-K2 × Gopt
Following a constant conductance
circle on the Smith Chart
PAE (%)
AM-PM = 6.3°
r ∆r
AM-AM Δr/r = 0.1 ≈ 1dB -20
15
r 10
θ
AM-PM θ = 0.1 rad ≈ 6° -30
5
I I
Gain Constellation -40 0
202 -10 -8 -6 -4 -2 0
Compression Rotation Power Backoff (dB)
EVMrms EVMrms
Lower = -28dB Higher = -24dB
average Pout average Pout
-31dBc -26dBc
-32dBc -32dBc -32dBc
Rg Cgd Rd
G D
VL Cjd
Vgs Cgs gmVgs Rds Cds
Rsub
AM-AM: gm, Rds, Cgs, Cgd, Cds
AM-PM: Cgs, Cgd, gm, Cds
Rs
Pout
gm (A/V), gm2 (A/V2), gm3 (A/V3)
)
22 Class-ABconv IMD3L w/o MGTR
2) 2
3 IMD3L_conventional_Doherty
(S/V
gm2 freq -15 IMD3H w/o MGTR
ggm3m3(S/V
IMD3H_conventional_Doherty
00 IMD3L w MGTR
IMD3L_MGTR_Doherty
2 gm3 -20
-2-2 IMD3H_MGTR_Doherty
IMD3H w MGTR
1 Conv -25
IMD3 (dBc)
-4-4
0 00 0.4
0.4 VVGS (V)
0.8
0.8 1.2
1.2 -30
GS (V)
gm3 impulse -35
-1
Sweet spot 44
Class-ABMGTR -40
Pout
-2
biasing 22
(S/V22))
freq -45
ggm3m3(S/V
-3 00
M1 M2
-50
0 0.3 0.6 0.9 1.2 -2-2 M3 M4
MGTR 6 8 10 12 14 16 18 20 22
VGS (V) -4-4 Output Power (dBm)
00 0.4
0.4
0.4 0.8
0.8
0.8 1.2
1.2
1.2
VVGS (V)
(V)
V GS (V)
GS
2 𝐶 2𝑊 𝐿 𝐶 _
𝐶 ≈ 𝑊𝐿 𝐶 _ 𝐶 ≈𝑊𝐿 𝐶 _ ≈
3 𝐶 3𝑊 𝐿 𝐶 _
2.6
L RL C
2.2
Capacitance (pF)
NMOS
C 1.8 PMOS
Cgd gd
NMOS+PMOS
1.4
Rs VG
Vin 1
Cgs
0.6
VPMOS 0 0.2 0.4 0.6 0.8 1 1.2
Vin (V)
IMD3(dBc)
IMD3
f0
Pin(dBm)
0<K<1 K>1
15
Gmax(dB)
Gain = 10dB Fmax=262(GHz)
10 Fmax=334(GHz)
(2) Unconditionally stable region (K>1):
5
Gain = 7.5dB
0
0 50 100 150 200 250 300 350 400
Dependent on device configurations: Frequency(GHz)
size/layout, biasing, matching
Limited fmax Limited device gain Gain @ 60GHz = 10dB 7.5dB
Fmax(GHz)
300 2.5um_finger
200
100
0
0.001 0.01 0.1 1
Device Layout Optimization and Extraction IDS/W(mA/µm)
400
PDK
Wf ↓ Nf↑ and Rg↓ 300 RC exraction
Fmax(GHz)
EM extraction
But Nf↑ other parasitics↑ 200
Gmax(dB)
20
Gmax vs. CN with a “volcano” shaped function 15
10 Fmax=334(GHz)
RF device (Class-A)
High input impedance 5 RF device Neutralized (Class-A)
0
Parasitics of CN are important at high mm-Wave 1 10 100 1000
Frequency(GHz)
20 2.5
Gmax
+Id -Id 2
Gmax (dB) 15 K-factor
K-Factor
+Vout -Vout 1.5
+Ifb 10
-Ifb 1
CN CN 5 0.5
+Vin -Vin
Cgd Cgd 0 0
0 10 20 30 40 50 60 70 80 90 100
CN (fF)
CS
R0 = 50Ω
LS CS = (1+QP2)/QP2
RL RL = RS
LS
RL = 12.5Ω
LS ω0 = 1/√ (LsCs)
CP Rs =
R0=50Ω R0/(1+QP2) CP
L-Match QP = R0ω0CP
RL
RL
Q=3 Q = 1.717 ... Q=3 Q = 1.07
CS
LP CS CS
R0=50Ω
LP LP R0=50Ω
L-Match
L-Match L-Match
R0 = 50Ω R0 = 50Ω
RL = 12.5Ω
RL = 5Ω
RL = 5Ω
Multi-Section L-Match
Lower Qnetwork and broader BW but with complexity, area, and loss
Hua Wang Fundamentals of RF and Mm-Wave 59 of 98
Power Amplifier Designs
Output Passive Network: Transformer
On-chip Transformers: Different compact lumped models [Long00].
P LP - M LS - M S P IP 1:n IS S
M IS•M12/LP LP LS IP•M21/LS
C0/n
Ideal Transformer Ideal Transformer
P Rp (1-km2)LP 1: (n/km) Rs S LS rs
P 1:n S
(1-k2)L/2 K:n
k2L/2 Cs Cpad
Output
Stage Z0
Cdev
k2L/2
(1-k2)L/2
ZOptimum Z5 Z4 Z3 Z2 Z1
R0 C1 L1 C3 L3 R0 R0 C1 L1 L3 C3 R0
(1) Splitting L2 +
Norton Transformation L2a′ L2b C2 (1) Δ-Y transformation L1′ L2′
R 0′ C1′ L1′ C3 L3 R0 R0 C1 L 3′ C3 R0
Even/Odd-Mode
Analysis [Ang03]
Imbalance(dB/degree)
-1 Passive loss -5 3
S-Parameters(dB)
Amplitude/Phase
S-Parameters(dB)
Broad Passband 2 Amplitude imbalance
-2 90-180 GHz -10 1
0
-3 -15 -1
-2
-4 Load Impedance Mismatch -20 -3
-4 Phase imbalance
-5 -25 -5
70 90 110 130 150 170 190 70 90 110 130 150 170 190
Frequency(GHz) Frequency(GHz)
𝟐𝒊×𝑷𝒉𝒂𝒔𝒆 (𝑺𝟐𝟏)
𝜞 𝑺 = 𝜞 𝑳𝒆
ZL Inverting
Balun
ZL Scaling
Balun
Polar PA Quadrature PA
Architecture Architecture
Output voltage swing ~ [0, VDD] Output voltage swing ~ [0, 2VDD]
Voltage-mode with low Zout, less coupling Current-mode with high Zout
ON: n cells 𝟐𝒏
𝑽𝒐𝒖𝒕 = 𝑽
N×ron/n OFF: N-n cells LS
+(n/N)VDD
𝝅 𝑵 𝑫𝑫
N-bit DAC t
Equivalent nC (N-n)C
-(n/N)VDD 𝟏 𝑽𝒐𝒖𝒕 𝟐 𝟐 𝒏 𝟐 𝑽𝑫𝑫 𝟐
Circuit RL = Ropt 𝑷𝒐𝒖𝒕 = = 𝟐( )
Ideal N×ron/(N-n)
𝟐 𝑹𝒐𝒑𝒕 𝝅 𝑵 𝑹𝒐𝒑𝒕
VDD
𝒏 𝑵−𝒏 𝑪
ZTH=Ron+1/jωNC
𝑷𝑺𝑪 = 𝑽𝑫𝑫 𝟐 𝒇𝟎
ON: n cells
OFF: N-n cells Ron nC LS +(n/N)VDD 𝑵
Thevenin t
Equivalent + -(n/N)VDD 𝑳𝒐𝒔𝒔 · 𝜷𝟐 𝑷𝒐𝒖𝒕 𝑹𝒐𝒑𝒕
Circuit
(n/N)VDD
- Series
RL = Ropt 𝜼= 𝒂𝒏𝒅 𝜷 =
Resonance 𝜷𝑷𝒐𝒖𝒕 + 𝑷𝑺𝑪 + 𝑷𝑺𝑾 𝑹𝒐𝒑𝒕 + 𝑹𝑺𝑾
New PA Architecture,
Linearization…
Linearization, Power
Combining, Reconfiguration …
Impedanceto Ropt
2 100
0º ZMain
RFin
Load Normalized
Impedance
Main 1.6 80
Main Aux. PA
Efficiency (%)
Inverter Tube Phase Z-Inverter
Tube 1.2 60
Compensation (λ/4 T-Line
Normalized
(λ/4 T-Line) with Z0 = Ropt)
ZMain′ 0.8 40
Zmain
-90º
IMain 0.4 Zaux 20
Main/Aux
Aux
PA Efficiency
IAux 0 0
Parallel Doherty PA Load ZL -20 -15 -10 -5 0
ZAux = Ropt/2 Output Power Back-off (dB)
Impedance to Ropt
ZMain 2 100
0º Zmain
RFin
Load Normalized
Impedance Main VMain 1.6 Zaux 80
Main Aux.
Inverter PA Efficiency
Efficiency (%)
Tube Tube Phase 1.2 60
Compensation
Normalized
(λ/4 T-Line) 0.8 40
-90º ZOut, Aux′ ZAux
Main/Aux
Aux VAux Load ZL 0.4 20
PA = 2×Ropt
0 0
Series Doherty PA Z-Inverter
ZOut, Aux (λ/4 T-Line -20 -15 -10 -5 0
with Z0 = Ropt) Output Power Back-Off (dB)
Ropt is the PA optimum load at 0dB PBO.
At Low Output Power (Pout <6dB PBO): Auxiliary PA is off with high Zout, Aux. Impedance
inverter ensures low Zout, Aux’, behaving like a voltage-mode amplifier.
At High Output Power (Pout is 6dB~0dB PBO): Both Main and Auxiliary PAs are turned on.
Auxiliary PA output increases ZMain = VMain×2Rload/(VMain + VAux)
RFin1 = A(t)
Acos(ωt+θ(t)+φ(t)) A
+φ(t)
PA1
Outphasing -φ(t)
Tube 1
Combiner
Tube 2 A
Outphasing θ(t)
Combiner
Network RFout =
RFin2 =
Acos(ωt+θ(t)-φ(t)) A(t)cos(ωt+θ(t))
PA2
Outphasing (Chireix) PA
*Outphasing efficiency based on ideal Class-
Key concepts and circuit elements B PAs and 30-degree Chireix compensation
High PA peak efficiency φ(t)=cos-1[A(t)/2A]
Backend overhead/complexity to generate outphasing signals (5×~7× BW expansion)
Often requiring extensive digital pre-distortion (DPD)
Outphasing network loss and carrier BW limiting
Hua Wang Fundamentals of RF and Mm-Wave 77 of 98
Power Amplifier Designs
PBO Efficiency Enhancement: Envelope-Tracking PA
Envelope-Tracking (ET) PA (L. Kahn 1952, A. Saleh 1983) [Kahn52] [Saleh83]
Constant Vdd ET supply voltage
PWM Switching Vdd
Supply Modulator Reduced
Power Loss
PA Efficiency
PA
RF Path Efficiency
EER/ET PA Enhancement
Unlike cascode
PAs, CK and VGK
are designed for
Vgate swings and Equal VDS
desired K×Ropt swings and
K×Ropt
C2 C2 C2
M2 M2 M2 Cshunt
Cgs2 Cgs2 Cgs2
Miller Effect:
(1-AV)× Cshunt
VG1 VG1 VG1
VMax VMax VMax
Vin C1 Vin C1 Vin C1
M1 M1 M1
Voltage-Mode SC Quadrature SC PA
ClassD-1 Digital PA Hybrid Digital Doherty Doherty PA [W. Yuan, RFIC15,
[D. Chowdhury, Digital Doherty PA JSSC16] Polar/Quadrature SC PA
Class-G PA [V. Vorapipat, RFIC16,
RFIC11, JSSC12] [S. Hu, RFIC14, [Y. Yin, ISSCC18,
[S. Hu, ISSCC15, JSSC16] JSSC17]
JSSC15] JSSC19]
Inductive
Tuning
41GHz Stacked-Device
A 60GHz Dual-Mode An E-band Parallel- Class-E SiGe PAs 90GHz Multi-Drive A 24-31GHz Class- A 28GHz Inductively
Class-AB PA Series Combiner PA [K. Datta, RFIC13, 45GHz Stacked-Device Stacked-Device PAs F/F-1 PA Degenerated PA
[D. Zhao, ESSCIRC12, [D. Zhao, ISSCC14, IMS13, JSSC14] Class-E CMOS PA [A. Agah, JSSC14] [S. Mortazavi, [S. Shakib,
JSSC13] T-MTT15] [A. Chakrabarti, ISSCC14] ISSCC16, JSSC16]
CICC12, T-MTT14]
Session 6
Paper 6.2 is a 4-Way Doherty Digital TX with 10dB Power Back-Off in the 5GHz band.
Session 26
Paper 26.1 is a 26-to-60GHz Coupler-Doherty PA for Over-An-Octave Back-Off Enhancement
Paper 26.2 is a Doherty-Like mm-Wave load-modulated balanced PA (LMBA) in 28nm CMOS
Paper 26.3 is a 30GHz dual-drive mm-Wave 5G PA for high efficiency
Paper 26.5 is a Watt-level quadrature switched/floated-capacitor digital PA with Back-Off
Enhancement
Paper 26.6 is a 5-to-6GHz current-mode subharmonic switching digital PA for Back-Off
Enhancement
Power amplifiers (PAs) have been the key building blocks for
wireless systems.
PA operation principles and PA classes
PA active device design, large-signal operation, and linearity
PA passive network design
Advanced PA architectures: Digital PAs, Back-Off Efficient PAs, and
Stacked PAs
RF/mm-Wave PA design examples and technology trends